view mp3lib/layer1.c @ 30884:53901d222e8a

Announce SMP support for Win32. Don't hardcode dwNumberOfProcessors=1 for Win32 anymore; the mutex/event code is still far from perfect, but now good enough that I can't find any codecs that breaks with this (tested on a quad with various codecs). This tells codecs they can use more than one core if they want to (some already did, by launching multiple threads even when told there was only a single core).
author sesse
date Wed, 17 Mar 2010 23:33:26 +0000
parents 32725ca88fed
children 0ad2da052b2e
line wrap: on
line source

/*
 * Mpeg Layer-1 audio decoder
 * --------------------------
 * copyright (c) 1995 by Michael Hipp, All rights reserved. See also 'README'
 * near unoptimzed ...
 *
 * may have a few bugs after last optimization ...
 *
 */

/*
 * Modified for use with MPlayer, for details see the changelog at
 * http://svn.mplayerhq.hu/mplayer/trunk/
 * $Id$
 *
 * The above-mentioned README file has the following to say about licensing:
 *
 *   COPYING: you may use this source under LGPL terms!
 */

#include "mpg123.h"

static void I_step_one(unsigned int balloc[], unsigned int scale_index[2][SBLIMIT],struct frame *fr)
{
  unsigned int *ba=balloc;
  unsigned int *sca = (unsigned int *) scale_index;

  if(fr->stereo == 2) {
    int i;
    int jsbound = fr->jsbound;
    for (i=0;i<jsbound;i++) {
      *ba++ = getbits(4);
      *ba++ = getbits(4);
    }
    for (i=jsbound;i<SBLIMIT;i++)
      *ba++ = getbits(4);

    ba = balloc;

    for (i=0;i<jsbound;i++) {
      if ((*ba++))
        *sca++ = getbits(6);
      if ((*ba++))
        *sca++ = getbits(6);
    }
    for (i=jsbound;i<SBLIMIT;i++)
      if ((*ba++)) {
        *sca++ =  getbits(6);
        *sca++ =  getbits(6);
      }
  }
  else {
    int i;
    for (i=0;i<SBLIMIT;i++)
      *ba++ = getbits(4);
    ba = balloc;
    for (i=0;i<SBLIMIT;i++)
      if ((*ba++))
        *sca++ = getbits(6);
  }
}

static void I_step_two(real fraction[2][SBLIMIT],unsigned int balloc[2*SBLIMIT],
	unsigned int scale_index[2][SBLIMIT],struct frame *fr)
{
  int i,n;
  int smpb[2*SBLIMIT]; /* values: 0-65535 */
  int *sample;
  register unsigned int *ba;
  register unsigned int *sca = (unsigned int *) scale_index;

  if(fr->stereo == 2) {
    int jsbound = fr->jsbound;
    register real *f0 = fraction[0];
    register real *f1 = fraction[1];
    ba = balloc;
    for (sample=smpb,i=0;i<jsbound;i++)  {
      if ((n = *ba++))
        *sample++ = getbits(n+1);
      if ((n = *ba++))
        *sample++ = getbits(n+1);
    }
    for (i=jsbound;i<SBLIMIT;i++)
      if ((n = *ba++))
        *sample++ = getbits(n+1);

    ba = balloc;
    for (sample=smpb,i=0;i<jsbound;i++) {
      if((n=*ba++))
        *f0++ = (real) ( ((-1)<<n) + (*sample++) + 1) * muls[n+1][*sca++];
      else
        *f0++ = 0.0;
      if((n=*ba++))
        *f1++ = (real) ( ((-1)<<n) + (*sample++) + 1) * muls[n+1][*sca++];
      else
        *f1++ = 0.0;
    }
    for (i=jsbound;i<SBLIMIT;i++) {
      if ((n=*ba++)) {
        real samp = ( ((-1)<<n) + (*sample++) + 1);
        *f0++ = samp * muls[n+1][*sca++];
        *f1++ = samp * muls[n+1][*sca++];
      }
      else
        *f0++ = *f1++ = 0.0;
    }
    for(i=fr->down_sample_sblimit;i<32;i++)
      fraction[0][i] = fraction[1][i] = 0.0;
  }
  else {
    register real *f0 = fraction[0];
    ba = balloc;
    for (sample=smpb,i=0;i<SBLIMIT;i++)
      if ((n = *ba++))
        *sample++ = getbits(n+1);
    ba = balloc;
    for (sample=smpb,i=0;i<SBLIMIT;i++) {
      if((n=*ba++))
        *f0++ = (real) ( ((-1)<<n) + (*sample++) + 1) * muls[n+1][*sca++];
      else
        *f0++ = 0.0;
    }
    for(i=fr->down_sample_sblimit;i<32;i++)
      fraction[0][i] = 0.0;
  }
}

static int do_layer1(struct frame *fr,int single)
{
  int clip=0;
  int i,stereo = fr->stereo;
  unsigned int balloc[2*SBLIMIT];
  unsigned int scale_index[2][SBLIMIT];
  DECLARE_ALIGNED(16, real, fraction[2][SBLIMIT]);
//  int single = fr->single;

//  printf("do_layer1(0x%02X 0x%02X 0x%02X 0x%02X 0x%02X 0x%02X 0x%02X 0x%02X )\n",
//    wordpointer[0],wordpointer[1],wordpointer[2],wordpointer[3],wordpointer[4],wordpointer[5],wordpointer[6],wordpointer[7]);

  fr->jsbound = (fr->mode == MPG_MD_JOINT_STEREO) ?
                         (fr->mode_ext<<2)+4 : 32;

  if(stereo == 1 || single == 3)
    single = 0;

  I_step_one(balloc,scale_index,fr);

  for (i=0;i<SCALE_BLOCK;i++)
  {
    I_step_two(fraction,balloc,scale_index,fr);

    if(single >= 0)
    {
      clip += (fr->synth_mono)( (real *) fraction[single],pcm_sample,&pcm_point);
    }
    else {
        int p1 = pcm_point;
        clip += (fr->synth)( (real *) fraction[0],0,pcm_sample,&p1);
        clip += (fr->synth)( (real *) fraction[1],1,pcm_sample,&pcm_point);
    }

  }

  return clip;
}